300 likes | 320 Views
. Can social network be used for location-aware recommendation?. Pasi Fränti , Karol Waga and Chaitanya Khurana. P. Fränti, K. Waga, and C. Khurana Can social network be used for location-aware recommendation
E N D
Can social network be used for location-aware recommendation? Pasi Fränti, Karol Waga and Chaitanya Khurana P. Fränti, K. Waga, and C. Khurana Can social network be used for location-aware recommendation Int. Conf. on Web Information Systems & Technologies (WEBIST'15), 558-565, 2015
Location-aware recommendation Location • Input: • User • Location • Time • Keyword (optional) Recommendations: • Nearby services • Photos of other users Press here Results
Four aspects of relevanceExample from practice 4. User and his network 1. Content • User profile • Social network • Text description • Keywords (tags) User: Pasi 2. Time • Recency of data • Season (not relevant in July) 3. Location • Distance to user Last skiing of winter Date: 4.4.2010 Location: N 62.63 E 29.86 Arppentie 5, Joensuu
Four aspects of relevanceExample from practice 4. User and his network 1. Content • User profile • Social network • Text description • Keywords (tags) User: Pasi 2. Time • Recency of data • Season (not relevant in July) 3. Location • Distance to user Last skiing of winter Date: 4.4.2010 Location: N 62.63 E 29.86 Arppentie 5, Joensuu
Scoring of services Search History Location Rating Rating by users in scale of 0-5 (SR) Keyword seached frequently (SF) Distance between user and the place (SL) Keyword searched nearby (SN) Normalizing to the scale [0,1] Keyword searched recently (SS) Keyword searched by the user (SU) Total score: S = NH + 2NL + NR + 1 History score: SH = SF + SN + SS + SU
Effectiveness of network • Social vs. information sharing • Buddy vs. stranger • Selected friends vs. ad hoc • On-line vs. offline network Popular networks:Facebook, Twitter, Google+, Instagram User activities: Likes, Comments, Retweet, favourite, rating. Activity stats in Facebook: 6 hrs/month, 2.7 billion likes/day
Small world phenomenon FB friends: Average=261 Total reach: world 17 M 4.6 B 261 68,000 • Entire world reachable in 6 steps (theory) • Experiment on Twitter users: 3.43 steps
Distribution of informationoptimistic Friends sharing Reached 0.01% 1% 0.1% 10% Total reach: 26 68 18 0 29,493 17,748 4,698 261 6,786 Efficiency reduces
Distribution of informationmore realistic Friends sharing Reached 0% 0.1% 0.01% 1% Total reach: 3 1 0 0 1,305 261 261 783 Efficiency reduces
Similarity of users Strength of link?
Methods for user similarity • Friendship in Facebook • Existing link similar • Friend of a friend not considered • Pages liked in Facebook • More matches more similar • Places visited in Mopsi • Visits same places similar
Similar Alice Bob Pages liked in Facebook Page similarity: Both like Hesburger Category similarity: Both like Fast Food Restaurants Page name Page category
Page similarity = 14%
Category similarity Mikko Book (2) Community (2) Attractions (1) Education (2) Travel (1) Community Organization (1) Company (1) Sports team (1) Amateur Sports team (1) Consulting (1) Business services (1) Radu Internet (1) Community organization (2) Tv show (1) Consulting (1) Media (1) Professional services (1) Education (4) Attractions (1) Website (1) Video game (1) Teacher (1) Non-profit organization (1) Sports event (1) Community (1) Health (1) = 22%
Pre-processing categories Media/News/Publishing→ Media TV channel→ TV Convert plural to singular Games/Toys→ Games Games→ Game Select first word
9 6 7 Location similarity visit statistics 0 0 0 Places 1 0 0 Visit frequencies 2 2 0 1 0 0 0 0 3 1 1 1 4 3 1 0 0 2
Similarity calculations Bhattacharyya distance 4 31 8 0.44 0.500.14 0.47 0.27 0.00 0.14 0.00 0.00 0.00 0.00 0.26 0.00 0.00 0.15 0.00 0.00 0.00 0.00 4 2 20 0.22 0.330.00 3 0 03 0.00 0.000.43 3 1 11 0.11 0.170.14 0 02 2 0.00 0.000.28 1 1 00 0.11 0.000.00 1 00 1 0.11 0.000.00 0 0 00 0.00 0.000.00 0.41 = 0.88 9 67 0.89 -ln = 0.13
Collected data • 293 places (Mopsi services) • User activities until 31.12.2014 • Photos taken • Tracking started or ended Joensuu sub-regionbounding box 63.44N 28.65E Municipalities: Joensuu Liperi Outokumpu Polvijärvi Kontiolahti Ilomantsi Juuka 62.25N 31.58E
Survey questions Q1: How similar you find the person is to you? Q2: How useful you find his/her Mopsi photos? Context for Q2: Does he recommend, via his/her Mopsi postings, useful and interesting places to visit in future.
User similarity Everyone is like Radu Not friends in Facebook Influential users
Expected usefulness • Mostly the same rankings (as with similarity) • Ranking of Pasi and Julinka improved • Expected vs. reality?
Most popular FB pages 2 University of Eastern Finland Joensuu This is Finland Stieg Larsson Phd, Masters and Postdoc Intern. Scholarships Joensuun Jääkarhut - Joensuu Polar Bears Joensuun Susi University of Eastern Finland (UEF) Vatakka Fotoaurinko Scientific Writing Assistant (SWAN) Carlson Ilosaarirock Festival Suomen Luonto House Sauna Jenni Vartiainen Official Hello Jessie Itä-Suomen yliopiston LUMA-keskus ABBA ABBA Facebook for Every Phone Hannes Hynönen - Fanisivu Jukolan viesti 10MILA The Herajärvenkierros Trail Kuopio Maraton 8 Impit Finland S+SSPR 2014 ECSE 7 Mopsi 6 Joensuu Science Park 5 UEF - School of Computing Odyssey 2014 4 SciFest Joensuu 3 Kaisa Mäkäräinen Jobs in Finland Joensuu - kaupunki idässä IMPDET-Le Polkujuoksu 13.9.2014
Page likes similarity • Correlates with user evaluations: • Similarity: 0.47 • Usefulness: 0.17
Similarity Graph page similarities Jukka 0.04 0.04 0.05 0.07 Oili 0.25 Radu Mikko 0.08 0.06 0.16 0.03 0.14 Rezaei Pasi 0.04 Chait 0.05 0.03 0.03 Julinka Andrei
Location data example • Correlates with user evaluations: • Similarity: 0.28 • Usefulness: 0.17
Conclusions • FB likes correlates to similarity • Location history has weaker correlation • Understanding of similarity interesting findings • • Answer: YES, but question remains HOW. To be continued…